Despite much recent research, there is still little systematic information about the phenomenology of panic attacks, and their possible causes remain obscure. We investigated panic attacks in the natural environment using an event sampling approach. Twenty-seven panic attack patients and 19 matched normal controls kept panic attack and self-exposure diaries for 6 days and wore an ambulatory heart rate/physical activity recorder for 3 days. Patients reported 175 attacks, generally of moderate severity. The most frequent symptoms were palpitations, dizziness/lightheadedness, dyspnea, nausea, sweating, and chest pain/discomfort. The results did not support the classification of panic attacks recently proposed by Sheehan and Sheehan, which requires three symptoms as a cutoff for panic attacks. Panic attacks classified by the patients as situational (i.e., occurring in feared situations) were more severe and occurred in situational contexts different from spontaneous attacks, but were otherwise phenomenologically similar. Heart rates did not change during spontaneous attacks and were only mildly elevated during situational attacks or during the 15 minutes preceding these attacks. These heart rate changes were interpretable as effects of anxiety, although physical activity showed a similar pattern of changes. Some normal control subjects reported on the panic diary primarily situational anxiety episodes that were phenomenologically similar to, albeit less severe than, the patients' episodes. Panic patients may sometimes fail to perceive environmental triggers for their attacks because many attacks classified as spontaneous occurred in classical "phobic" situations. Furthermore, the comparison of concurrent diary and retrospective interview and questionnaire descriptions showed that panic patients have a tendency toward retrospective exaggeration. Implications for the assessment, definition, and classification of panic attacks are discussed.